♻️ Compatibility with vllm main

[prashantgupta24]

Changes

• Remove prompt adapter config

  • Based on upstream vllm changes in [V0 Deprecation] Remove Prompt Adapters vllm#20588

• Implement get_supported_tasks in model_runner for online API

  • only needed for online API, related PR - [V1] Get supported tasks from model runner instead of model config vllm#21585

  Implementation for online API:

  if envs.VLLM_USE_V1:
      supported_tasks = await engine_client \
          .get_supported_tasks()  # type: ignore
  else:
      supported_tasks = model_config.supported_tasks
  

Related Issues

fix #336

[github-actions]

👋 Hi! Thank you for contributing to vLLM support on Spyre.
Just a reminder: Make sure that your code passes all the linting checks, otherwise your PR won't be able to be merged. To do so, first install the linting requirements, then run format.sh and commit the changes. This can be done with uv directly:

uv sync --frozen --group lint --active --inexact

Or this can be done with pip:

uv pip compile --group lint > requirements-lint.txt
pip install -r requirements-lint.txt
bash format.sh

Now you are good to go 🚀

[maxdebayser]

The pooling task changes are already part of the embedding PR. There are more upstream changes around the pooling tasks and I think they should be part of a follow-up PR. I don't want to pile up more changes in a PR that is already big enough.

[prashantgupta24]

Yeah I was trying to get a minimum set of changes in so that the tests pass against vllm:main. I think the pooling tasks are non-breaking, I need to get get_supported_tasks in for vllm:main to work atm

[prashantgupta24]

@maxdebayser I think just need this piece of code from your PR to enable vllm:main breaking changes, I've removed all other pooling changes from this PR. I'm fine in waiting for your PR to get in first in which case I'll rebase and remove this change

[prashantgupta24]

Ideally we would want this to be coming from the model directly:

if is_text_generation_model(model):
            supported_tasks.append("generate")

but SpyreCausalLM doesn't seem to support it atm

type(self.model)
<class 'vllm_spyre.model_executor.model_loader.spyre.SpyreCausalLM'>

is_text_generation_model(self.model)
False

[yannicks1]

lgtm! thanks for taking that on!